Introduction

Hands-on experiments in the classroom are key for conveying knowledge of complex natural phenomena that might otherwise remain inaccessible to students (Kind 2015). Yet, for many years, a clear focus on instructional and theoretical teaching has dominated the educational landscape. Today, new national educational standards (KMK 2005) replace these long-established methods and raise hope of more practical approaches to science teaching. However, one must bear in mind that some schools may lack the financial means to build and stock laboratories (e.g., Raviv et al. 2019) which may require expensive and specialist equipment.

For schools without the necessary means, outreach laboratories (e.g., at universities) offer opportunities for students to experience science in authentic settings. Many such laboratories have been developed over the last two decades and offer opportunities to foster students’ interest and motivation (Glowinski and Bayrhuber 2011). Hereby, solution-oriented approaches to scientific problems and an adequate learning environment are important (e.g., Nasir et al. 2006, Brody et al. 2007). Our outreach program, Simply inGEN(E)ious! DNA as a carrier of genetic information, meets these requirements: Designed as a 1-day-long module, it involves hands-on experiments, prediction of potential results, and a modeling phase (Mierdel and Bogner 2019a).

In our previous study, we compared two modeling variants: model viewers and modelers. Model viewers worked “with a commercially available school model of DNA structure” while modelers “were required to generate a DNA model using assorted handcrafting materials” (Mierdel and Bogner 2019a, p. 1). Most of the existing literature suggests that hands-on modeling contributes to cognitive achievement (e.g., Jackson et al. 2008, Passmore et al. 2009), yet we found that model viewers had a higher mid-term increase in knowledge. Thus, “we dispensed a more detailed model evaluation for modellers” (Mierdel and Bogner 2019a, p. 12). To fill this gap in our research, we selected two variants related to students’ model evaluation: modellers-1, who only participated in one evaluation phase, and modellers-2, who additionally completed a second evaluation phase. We based the latter on our model-viewing approach.

Our study aims to observe the potential influences of both model evaluation variants on cognitive achievement. We first outline the relevant theoretical background, including knowledge of learning and model learning which influenced the way we developed, conducted, and evaluated our intervention. We then provide a brief explanation of the phases of our intervention before we present our research questions.

Theoretical Background

Learning

Despite numerous suggestions, a universally accepted definition of “learning” has not yet been developed (e.g., Domjan 2015, Haselgrove 2016). Learning is often summarized as “a change in behavior [depending on] experience” (De Houwer et al., 2011, p. 631, Thorpe 1943), a view which is supported by constructivists (Hodson 1998; Lachman 1997). Thereby, students autonomously verify or falsify hypotheses and modify preconceptions accordingly or develop new ideas (Schwarz et al. 2009). These competencies also reflect (Kolb’s 1984, Abdulwahed and Nagy 2013) categorization of learners’ abilities: concrete experience ability, reflective observation ability, abstract conceptualization ability, and active experimentation ability. Hence, the construction of knowledge from chunks of information requires experiential processes to transform such information into a coherent mental model.

Real experiments in class or outreach settings provide the necessary basis for students to build up knowledge and allow them to manage scientific equipment. Such first-hand experience is vital to understanding the nature of the physical world and science, where experimental outcomes are not always as expected (Loveys and Riggs 2019; Olympiou and Zacharia 2012). Solomon (1980, p.13) even described hands-on experimenting as key to increasing scientific knowledge: “Science teaching must take place in the laboratory; science simply belongs there as naturally as cooking belongs in the kitchen and gardening in the garden.” Different levels of scientific knowledge can, thus, be achieved by combining hands-on experimentation with the development of scientific thinking skills. This approach significantly enhances skills set by the national standards (KMK 2005), such as the understanding of scientific processes. Together with an open and flexible learning environment, critical thinking skills and self-evaluation may be fostered. This is especially important when it comes to reassessing results. An effective method for encouraging self-evaluation is reflective writing (Kovanović et al. 2018) about particular experimental phases. This approach can encourage students to rethink and reassess certain steps in the experimental process (Scharfenberg and Bogner 2013; for details, see below). In this “self-directive process,” students “transform their mental abilities into academic skills” (Zimmerman, 2002, p. 65).

Model Learning

Models are tools commonly used to visualize and explain phenomena (Krajcik and Merritt 2012). Windschitl and Thompson (2006, p. 796), for instance, described models as “hypothesized relationships among objects, processes, and events”. Thereby, models can reveal underlying mechanisms, show causal links, raise questions, and test multiple hypotheses. Whenever a prediction proves incorrect or new evidence emerges, a model can be adapted and refined (e.g., Passmore et al. 2009, Carpenter et al. 2018). Models are regarded as the cornerstones of every scientific discipline, combining theory with modeled processes and functions to explain how things really are (Ates and Erylmaz 2011). Thus, modeling in science classes is invaluable for student learning (e.g., Jackson et al. 2008, Passmore et al. 2009). Although natural phenomena are often difficult to observe, models can provide an authentic alternative experience, are often easier to understand, and do not require exhaustive preparatory work (e.g., Windschitl and Thompson 2006, Carpenter et al. 2018). Students, interalia, gained better cognitive achievement scores when applying three-dimensional DNA models rather than two-dimensional DNA models (e.g., Rotbain et al. 2006, Saka et al. 2006).

Modeling may lead to more productive student activity in the classroom, emphasizing vital scientific practices, including the need to “engage in inquiry other than controlled experiments, [to] use existing models in their inquiries, [to] engage in inquiry that leads to revised models, [to] use models to construct explanations, [to] use models to ‘unify’ their understanding, and [to] engage in argumentation” (Passmore et al. 2009, p. 397). We decided to include these activities in our three-dimensional modeling approach (for details, see below).

Intervention Phases

We divided our 1-day module into four different intervention phases (Table 1).

Table 1 Quasi-experimental study design
Full size table

Pre-lab Phase

As many students are not familiar with laboratory equipment or scientific work, we provide an appropriate pre-lab phase. This phase addresses three scientific aspects, in particular: affective dimensions, introduction to laboratory techniques, and introduction to theoretical concepts (e.g., Sarmouk et al. 2019). Students, thus, learn how to handle the equipment necessary for experimentation, such as micropipettes and centrifuges. Teachers act as external guides and introduce theoretical concepts via presentations and demonstrations, which are key to understanding the subsequent experiments. This phase should also prevent students from feeling overwhelmed in response to the experimental situation (e.g. Kalyuga 2009).

DNA-Related Theoretical and Experimental Phases

Theoretical Phases

Having provided a real-life background to the subsequent experiments (DNA relevance; Table 1), the underlying scientific basics of DNA isolation and gel electrophoresis were introduced. Students were then encouraged to connect new information with prior knowledge and hypothesize about the experiments’ potential outcomes (Mierdel and Bogner 2019a). Our decision to provide information in a series of theoretical phases ensures that students are not overwhelmed by too much new information at once and allows them to focus on the experimental phase ahead. This is particularly important when explaining the difficult concept of gel electrophoresis which not only involves chemical knowledge about the DNA’s contents but also abstract thinking in order to imagine the processes involved.

Experimental Phases

Ours is an evidence-based, two-step approach (Scharfenberg and Bogner 2013), in which students answer questions in their individual workbooks and think about subsequent experimental procedures. They work in pairs to discuss every step before carrying out experiments that effectively combine hands-on and minds-on activities and require students to do more than simply follow instructions (Scharfenberg and Bogner 2013).

Model-Related Phases

Both model-related phases directly followed the experimental, DNA-related phases. As we regard models as vital to visualizing and explaining phenomena in science and science education (Krajcik & Merritt 2012), we subdivided our model-related phases into a mental modeling phase involving text analysis, a modeling phase involving craft materials, a model evaluation-1 phase, and a model evaluation-2 phase. Only modellers-2 participated in all four phases.

As in our previous study, we based our model-related phases on the four main stages of the Model of Modelling (Justi and Gilbert 2002, p. 370 ff.): (1) “experience[s] of the phenomenon being modelled”, (2) “forming a mental model”, (3) “decision … about the mode of representation in which it is to be expressed”, and (4) “testing … scope and limitations of the model”. Mental modeling is, thereby, key to providing a theoretical basis for experimental findings. According to Franco and Colinvaux (2000), building mental models involves reasoning about previously obtained knowledge to make predictions and derive new ideas from it. A text about the discovery of the DNA’s structure (Usher 2013), thus, provides fundamental knowledge about the necessary components of the DNA, which will be applied to the students’ simplified mental model (e.g., Mierdel and Bogner 2019b, Franco & Colinvaux 2000).

After determining the DNA’s representation and building the handcrafted model, model evaluation-1 phase was organized as a reciprocal self-evaluation. As combining sketching and handcrafting models proved to be important (e.g., Prabha 2016, Orhan and Sahin 2018), modellers-1 and modellers-2 evaluated their handcrafted DNA model based on their earlier paper-and-pencil version. Another effective method for encouraging self-evaluation is reflective writing (Kovanović et al. 2018). Open-ended questions about model-related components encourage students to rethink and reassess certain steps and decisions in their development of the mental model into its physical counterpart (Mierdel and Bogner 2019a). Modellers-2 also assessed their handcrafted DNA models using a comparison-based self-evaluation with a commercially available DNA demonstration model. In other words, we based this second evaluation phase on our previous model-viewing approach.

Interpretation Phase

Here, students compared their hypothesis about the outcome of their experiments with the gel electrophoresis’ images. They also discussed their individual models in class and compared these to the molecular DNA model by Watson and Crick which, in most cases, differed from the students’ models (Mierdel and Bogner 2019b).

Objectives of the Study

The present study tries to answer the following research questions:

  1. 1

    How does the application of one or two model evaluation phases influence overall cognitive achievement and, in particular, the model-related knowledge developed during the hands-on laboratory?

  2. 2

    How does the additional evaluation-2 phase of modellers-2’s three-dimensional models influence their overall learning?

Thus, we had three specific objectives:

  • to assess students’ overall cognitive achievement and the model-related knowledge of modellers-2, which we would compare with that of modellers-1

  • to determine the quality of the evaluation-2 phase and the DNA component that modellers-2 correctly identified in their handcrafted models

  • to examine the potential correlations between modellers-2’s performance in the model evaluation-2 phase and modellers-2’s cognitive achievement

Materials and Methods

After a brief introduction to our educational intervention, its design, and the independent variable, we explain the modeling phases and the model evaluation phases. We then describe the students’ sample, discuss the dependent variables, and outline the statistical methods applied.

Educational Intervention, Design, and Independent Variable

The 1-day, hands-on module offered inquiry-based learning activities focused on the structure of DNA, aimed at ninth graders. Students worked in pairs to complete their tasks with guidance provided in a workbook (for a detailed module description, see Mierdel and Bogner 2019a). The content of the module is in line with the state’s syllabus and follows the national competency requirements (KMK 2005).

We conducted two versions of the intervention which differed concerning students’ evaluations of their models. Since often only quasi-experimental designs are feasible for students in intact class groups (Cook and Campell 1979), student classes were randomly assigned to each of the evaluation variants. The students were, thus, divided into modellers-1 and modellers-2, respectively, as the independent variable (Table 1).

Both versions of the intervention began with a pre-lab phase (50 min) wherein students were familiarized with the lab equipment and relevant working techniques. Thereafter, the criminological relevance of DNA was introduced to contextualize the two main DNA-related experimental phases: DNA isolation from oral mucosal cells (60 min) and agarose gel electrophoresis (85 min). Both were connected to model-related phases (60 min). Students then tried to retrace Watson and Crick’s research to solve the molecular puzzle of the DNA’s structure.

The Modeling Phases

These phases (Table 1) were key in providing a theoretical basis for experimental findings. Having read about the discovery of the structure of DNA (Usher 2013), students discussed and answered questions in their workbooks (e.g., “DNA’s backbone: Label its components and describe their set up”; for details, see Electronic Supplementary Material [ESM] 1 as an online resource). The aim was to enable the students to internalize key aspects of the text which would later allow them to mentally model the DNA’s structure. To this end, the text included references to all key components of the DNA (e.g., base pairing).

In the next stage, students transformed their mental DNA models into physical, handcrafted DNA models (Table 1) using a DNA-modeling kit containing crafting materials (e.g., colored beads and pipe cleaners; for three model examples; Table 2, 1st column).

Table 2 Assessment of modellers-2’s evaluation phases
Full size table

The Model Evaluation Phases

These phases (Table 2, 2nd and 3rd columns) were organized as a reciprocal self-evaluation of students’ models. Modellers-1 and modellers-2 all evaluated their handcrafted DNA model against their paper-and-pencil sketch of the model (evaluation-1). Modellers-2 then conducted an additional, comparison-based assessment of their handcrafted DNA models, in which they were asked to draw a paper-and-pencil model with labels to identify the model’s components (Table 2, 2nd column) and the elements of their handcrafted models. This was an opportunity to reflect on the process of building the model and on its informative value. In the meantime, ideas about the accuracy of their models could be exchanged. While modeling, students might have become aware of differences between their own and their classmates’ models. Once they had completed a sketch of their models, students answered two open-ended questions on their worksheets: “Which features of the original DNA molecule are simplified in your model” and “Explain why one might create different models of one biological original (in our case, the structure of the DNA)?” Thus, students were required to consider the scope and limitations of their models.

Modellers-2 also completed evaluation-2 (Table 1), wherein they were asked to compare and contrast their handmade models with a commercially available DNA model. A self-evaluation sheet (ESM 2) displaying the image of this scientific model served as the basis of their assessments. Each component included in both models was tagged to assist students’ self-evaluations of their models.

In both versions, students’ findings from the model phase were integrated into a final interpretation phase. Here, students discussed experimental results of the gel electrophoresis, which they compared with previous hypotheses (Table 1).

Participants

Altogether, 296 ninth-graders (higher secondary school) participated in our study (girls 52.0%, boys 48.0%; MClass size = 22.8, SD = 6.2; MAge = 14.6, SD = 0.8). Six classes took part as modellers-1 (n = 151) and seven classes as modellers-2 (n = 145). Modellers-1 teamed up in 77 groups (75 2-person groups and two students working individually due to illness), modellers-2 in 73 groups (72 2-person groups and one student working individually due to illness). To avoid bias, we compared students’ prior knowledge of biology with the respective biology grades. We found no significant difference (Mann-Whitney U test [MWU]: Z = − 1.144, p = .253). Moreover, we compared individual students’ prior in-class experience of modeling (3-item scale, adapted from Authors 2007; Cronbach’s Alpha .62) and did not discover a significant difference (MWU: Z = −0.859, p = .390).

Participation was voluntary. Written parental consent was given prior to students’ participation in our study, although the data collection was pseudo-anonymous and students could not be identified. The study was designed in accordance with the Declaration of Helsinki (2013), and the state ministry approved the questionnaires used.

Dependent Variables

As dependent variables, we examined students’ knowledge in a repeated measurement design: a pre-test (T0) 2 weeks before the intervention, a post-test (T1) after the module, and a retention test 6 weeks thereafter (T2). We examined students’ sketches and their responses to the open questions from the evaluation-1 phase; for modellers-2, we additionally assessed the evaluation-2 phase. Throughout the entire intervention, students were unaware of any testing schedules.

Students’ Knowledge

We applied an ad hoc knowledge test comprising 30 multiple-choice items: 12 items (examples; Table 3) assessed knowledge of the DNA-related phases (DNA relevance, hands-on isolation, and gel electrophoresis of DNA; Table 1) and 18 items (Table 3) analyzed knowledge related to the model phases (Table 1).

Table 3 Knowledge item examples related to the DNA-related phases (A) and the modeling phase (B)
Full size table

Content validity was given as the items were consistent with the state syllabus. Regarding construct validity, inter-item correlations below 0.20 (T0 = 0.08; T1 = 0.19; T2 = 0.18) confirmed that each item referred to different knowledge facets. Furthermore, the items’ heterogeneity concerning complex constructs, such as cognitive achievement, emphasizes the given construct validity (Rost 2004). Cronbach’s alpha values of 0.71 (T0), 0.64 (T1), and 0.70 (T2) indicate acceptable internal consistency when the value exceeds 0.70. According to Lienert and Raatz (1998), values between 0.50 and 0.70 allow for the differentiation of groups. Item difficulties (percentage of correct answers, Bortz and Döring 1995) ranged between 7% (high difficulty) and 88% (low difficulty). In the case of our intervention, the item difficulties reversed from T0 to T1 (Fig. 1) and generally decreased.

Fig. 1
figure 1

a Repeated measurement design of both instructional variants in the outreach lab. b Item-difficulties for T0 and T1. DNA-related items shortened “D”, model-related items shortened “M”. Item examples shown in Table 3 highlighted in light grey (note the shift between the pre- and post-schedule)

Full size image

We calculated the students’ scores and analyzed these with regard to increases in knowledge (T1 minus T0) and retention rate (T2 minus T0). However, these different variables do not reflect actual knowledge growth. Thus, we calculated the actual learning success with respect to the maximal attainable score (30 correct answers): (T1 − T0) × (T1/30) and the persistent learning success (T2 − T0) × (T2/30) (Scharfenberg et al. 2007). Increased knowledge is, hence, weighted according to the students’ actual knowledge, making it possible to compare cognitive achievement despite some students exhibiting a huge increase in knowledge yet low final scores, and vice versa.

We also calculated correlations between biology grades and post-test (T1) scores for overall and model-related knowledge items using Spearman-Rho (Field 2012).

Evaluation-1 Phase

In order to compare the evaluation variants for the evaluation-1 phase, we assessed students’ model sketches (changed after Langheinrich and Bogner 2015; for definitions, examples, and frequencies, see ESM 3). We randomly selected 26 out of 150 drawings for a second scoring (17.3%). Cohen’s kappa coefficient (Cohen 1968) scores of 0.88 and 0.82 for intra-rater and inter-rater reliability showed an “almost perfect” rating (Wolf 1997, p. 964). To avoid bias, sketches of both variants were compared and no significant difference could be identified (MWU: Z = − 0.745, p = .456).

Using content analysis (Bos and Tarnai 1999), we iteratively categorized the statements that students made in response to the open questions. For the first question “Which features of the original DNA molecule are simplified in your model”, four categories were employed: level of DNA, level of substance, level of particles, and level of structure (for definitions, examples, and frequencies, see ESM 4). We randomly selected 58 out of 384 statements for a second scoring (15.1%). We computed Cohen’s kappa coefficient (Cohen 1968) scores of 0.98 and 0.78 for intra-rater and inter-rater reliability, which showed a “substantial” to almost perfect rating (Wolf 1997, p. 964). For the second question, “Explain why one might create different models of one biological original (in our case, the structure of the DNA)?”, we applied the adapted category system of Mierdel and Bogner (2019b) and identified five categories: individuality of DNA, different interpretation, different model design, different focus, and different research state (for definitions, examples, and frequencies, see ESM 5). We randomly selected 27 out of 150 statements for a second scoring (18.0%). We computed Cohen’s kappa coefficient (Cohen, 1968) scores of 0.75 and 0.70 for intra-rater and inter-rater reliability, which showed a substantial rating (Wolf 1997, p. 964). To avoid bias, we compared the two evaluation variants in terms of category frequencies of responses to both open questions. We did not find any significant contingencies (adjusted Pearson’s C ≤ .192; p ≥ .065; Pearson 1900).

Evaluation-2 Phase

A three-step approach was applied to evaluate the evaluation-2 phase (Table 4):

  • Documentation of the students’ self-evaluation: We counted each box that students had tagged on their self-evaluation sheet as one point (maximal score 14 points)

  • Assessment of students’ self-evaluation sheets: We analyzed the tagged boxes’ conformity on the self-evaluation sheets using the respective models. Appropriate tags received one point each. If students tagged all their boxes correctly, they would reach the maximal core as was recorded on their self-evaluation sheets

  • Assessment of students’ models: We independently assessed the models. Correct features each received one point whether or not they had been identified by the students (maximal score 14 points)

    Table 4 Assessment of the modellers-2 evaluation-2 phase
    Full size table

A comparison between the documented boxes and the assessment of the self-evaluation sheets enabled us to determine the extent to which students had correctly evaluated their models. Lower scores on the self-evaluation sheet indicate students’ mistakes when assessing the quality of their models. However, a lower model score also indicates that a student may have documented model features that were not given. By contrast, higher model scores indicate model features that the student did not identify as such.

Statistical Analysis

We applied nonparametric methods due to an abnormal distribution of variables (the Kolmogorov-Smirnov test (Lilliefors modification): partially p < .001), and, consequently, use boxplots to illustrate our results. Intra-group differences over the three test dates were analyzed using the Friedman test (F) in combination with a pairwise analysis from T0 to T1 and T2, and from T1 to T2, using the Wilcoxon (W) signed-rank test. The Mann-Whitney U tests (MWU) were used to evaluate inter-group differences. Due to multiple testing, we applied a Bonferroni correction (Field 2012). In the case of significant results, effect sizes r (Lipsey and Wilson 2001) were calculated with small (> 0.1), medium (> 0.3), and large (> 0.5) effect sizes. For correlation analyses, we applied Spearman’s rank correlations and report Spearman’s Rho values.

Results

We first provide an overview of our intra-group and inter-group analyses with regard to overall and model-related knowledge. This is followed by a detailed assessment of the evaluation-2 phase.

Intra-Group Analyses of Cognitive Achievement

Intra-group analysis (F and W tests; Table 5) revealed significant changes for modellers-1 and modellers-2, in terms of both overall and model-related knowledge: They initially increased their knowledge at both levels, which then dropped between T1 and T2, but not below prior levels (T0). This suggests that students gain short-term and mid-term knowledge throughout the intervention (Table 6).

Table 5 Cognitive achievement of the students’ sample as a whole and model understanding items only
Full size table
Table 6 Dependent variables for both modellers-1 and modellers-2, analyzed with regard to knowledge scores, difference variables, and learning success. Overall knowledge items and model-related items only were differentiated
Full size table

Inter-Group Analyses of Cognitive Achievement

To account for differences in students’ prior knowledge (Table 6, superscript a), we calculated difference variables for short-term increases in knowledge and mid-term retention rates and learning success variables to assess inter-group differences. These were the only variables taken into account. Based on sum scores, increases in knowledge (T1-T0) and retention rate (T2-T0) were calculated (Field 2012) for overall (30 items) and model-related (18 items) knowledge scores (Table 6).

Overall Knowledge

Scores of the overall knowledge test, which included both DNA-related and model-related knowledge items, showed differences in increases in knowledge and retention rates between modellers-1 and modellers-2. Modellers-2 scored significantly higher in increased knowledge and retention rate than modellers-1, with a medium-to-large effect size (Table 6, superscripts d/e).

As different variables do not display the actual cognitive achievement, we analyzed learning success variables (see above). For short-term actual and mid-term persistent learning success, significant differences in overall knowledge were identified with medium-to-large effect sizes (Table 6; superscripts h/i). Compared to modellers-1, modellers-2 achieved higher scores in terms of actual and persistent learning success (Fig. 2).

Fig. 2
figure 2

Changes in overall knowledge scores for both modellers-1 and modellers-2. All participating groups increased their actual learning success scores: (T1 – T0) x (T1/30). Persistent learning success scores: (T2 – T0) x (T2/30), however, dropped in comparison to actual learning success scores

Full size image

Model-Related Knowledge

Regarding scores based on 18, model-related knowledge items, modellers-2 achieved significantly higher increases in knowledge and retention rates with a medium effect size (Table 6, superscripts d/e). At the level of learning success, modellers-2 achieved higher scores (Fig. 3). However, model-related cognitive achievement exceeded that of overall knowledge (Table 6, superscripts j).

Fig. 3
figure 3

Changes in all 18 model-related knowledge items for both modellers-1 and modellers-2. All participating groups increased their actual learning success scores: (T1 – T0) x (T1/18). Persistent learning success scores: (T2 – T0) x (T2/18), however, dropped in comparison to actual learning success scores

Full size image

Correlation Biology Grades

Biology grades and post-test (T1) scores displayed a weak negative correlation (rS = − .209, p < .001). We obtained similar results by correlating students’ biology grades and post-tests for model-related knowledge (rS = − .202, p = .001). Splitting our evaluation variants, modellers-1 did not reveal any negative correlations (p ≥ .190); modellers-2, however, displayed significant negative correlations of rS = − .235 (p = .006) for overall and of rS = − .196 (p = .026) for model-related knowledge. Thus, students with lower grades tend to achieve higher scores in overall and model-related knowledge post-tests than students with better grades. This was particularly evident in our evaluation-2 variant.

Assessment of Evaluation-2 Phase

Intra-group analyses of modellers-2 revealed the differences between students’ self-evaluations and our assessment of their self-evaluation sheets and their model (Table 4; F: chi-square 59.531, df = 2; p < .001). Pairwise analysis revealed lower scores for their assessed self-evaluation sheets with a large effect (W: Z = − 6.220, p < .001; r = .728). Thus, students identified as correct features that were not given in their model. This discrepancy was evident across all analyses sectors (Table 4; F: chi-square ≥ 16.919, df = 2, p < .001, in each case; W: Z ≤ − 2.715, p ≤ .007, in each case; r ≥ .318). In contrast—and also to a significant effect—some of the assessed models scored higher than the assessed self-evaluation sheets (W: Z = − 5.804, p < .001; r = .684). Thereby, the students did not identify all of the correctly modeled features. This phenomenon was also evident across all analyses sectors (Table 4; W: Z ≤ − 3.938, p ≤ .001, in each case; r ≥ .464).

Correlation analysis revealed small-to-medium correlations between the models’ assessment scores, reflecting the models’ quality, the actual learning success (rS = .260, p = .003), and the persistent learning success (rS = .251, p = .005). Further correlations between the abovementioned variable and model-related actual learning success (rS = .215, p = .015) and the model-related persistent learning success (rS = .218, p = .014) were of particular importance. Thereby, models’ assessment scores showed a small correlation with students’ prior in-class experience of modeling (rS = .217, p = .014).

Discussion

Our aim was to examine potential differences between participation in the two evaluation variants. The data suggest that students’ short-term and mid-term DNA- and model-related knowledge was improved by the additional evaluation-2 phase.

Cognitive Achievement

Overall Knowledge

The effects of an additional model evaluation-2 phase on the actual and persistent learning success indicate that this is a positive approach that supports effective learning. According to Ainsworth (2008), modeling affects three dimensions of learning which enable students to abstract, extend, and relate their knowledge; to identify familiar concepts; and to appropriately display multiple aspects of the respective phenomenon. Altogether, this may lead to a deeper understanding of the subject (Oh and Oh, 2011). In our previous study, the model evaluation-1 phase had already suggested that a 1-day intervention combining model-related activities and hands-on experimentation could encourage learning in the abstract field of molecular genetics (Mierdel and Bogner 2019a). The effective integration of difficult scientific theory and working techniques into biology classrooms is, thus, a promising approach, as has been argued by Peel et al. (2019). Still, there was room for improvement which is why we conducted our present intervention with the additional model evaluation-2 phase. Compared to evaluation-1 variant (modellers-1), those participating in evaluation-2 (modellers-2) achieved higher scores regarding increases in knowledge and retention rate and received higher actual and persistent learning success scores (Table 6). Students not only developed an initial sketch of their mental DNA model but also handcrafted a three-dimensional model which they assessed in a comparison-based self-evaluation during which they worked in pairs to discuss and review their approach. In science, this type of evaluation is vital to assess the adequacy of previously developed models and make efforts to improve upon them. In our study, these discussions required students to focus on providing consistent explanations and balanced assessments of their models (e.g., Passmore et al. 2009, Schwarz et al. 2009).

Model-Related Knowledge

Specific analysis of model-related knowledge items revealed a higher increase in knowledge and retention rate, for modellers-2 than for modellers-1. It also revealed higher actual and persistent learning success scores for modellers-2 (Table 6). Thus, our additional evaluation phase seems to impact model-based knowledge items. Schwarz et al. (2009) and Bryce et al. (2016) suggest a link between modeling practices in the classroom and students’ learning success. Our measured cognitive achievement in model-related items (Table 6) might, then, be due to a “progression in knowledge and skills required for modelling, necessarily [entailing] progression in knowledge about the nature of models” (Gilbert and Justi 2016, p. 195). This possibility is also reflected in correlation analysis, with small-to-medium correlations between models’ assessment scores and actual as well as persistent learning success. Thus, we can conclude that understanding of models is connected to knowledge about models (Peel et al. 2019).

Modeling also helped students to organize information about DNA as a model. It ultimately contributed to a deeper understanding of the learning content (e.g., Bryce et al. 2016; Grünkorn et al. 2014) and, in our case, supported students’ understanding of DNA as a model.

Evaluation-2 Phase

Although actual and persistent learning success differed in both evaluation variants, there were also quantifiable differences in modellers-2’s handcrafted models (for example, Table 2). Some students correctly identified and labeled the DNA’s different components, while some students only identified a few and others only modeled its basic structure. This result is in line with Howell et al. (2019) and Kim et al. (2015), who described students’ difficulties in understanding DNA’s structure-function relationships. Other studies about scientific modeling do not focus on DNA but, for instance, on the acoustic properties of materials (Hernández et al. 2015), natural selection and antibiotic resistance (Peel et al. 2019), or general classroom examples in biology, geography, and physics (Schwarz et al. 2009). It could also be discussed that students were, to a certain extent, unable to transfer their previously obtained factual knowledge into the new form of a model. Yet, although hardly any of the students had sufficient prior experience, they nonetheless found a way to excel—an achievement particularly notable among those who were considered, grade-wise, to be low achievers (Bamberger & Davis 2011). Our own prior research suggests that most students find models (and other types of images: charts, graphs, diagrams, etc.) very difficult to engage with (personal research). Moreover, our results are in line with the findings of Quigley et al. (2017) who used the EcoSurvey tool to assess different kinds of models across different classrooms and correlated these with learning success. The authors suspected differences in modeling experience to be the underlying cause of success or failure. We also found a small correlation between the students’ model quality and their prior experience in modeling at school. In our case, individual students stood out from the crowd if they either had a better understanding of models, processed new information more effectively, or read the text more carefully. Analysis of answers to DNA model–related questions on the students’ worksheet from model evaluation-1 (Table 1) supports this approach (ESM 4, ESM 5). Sample answers showed a broad understanding of DNA as a model; for instance, at particle level, that “the sugar and phosphate molecules are simplified (they usually consist of various atoms)” (ESM 4). This supports the claim that modeling is closely connected to students’ academic performance (Quigley et al. 2017), as long as the modeling is not only about “doing school” but has scientific relevance (Schwarz et al. 2009, p.652).

Methodological Aspects

Firstly, as noted by Hernández et al. (2015, p. 257), several “recurring cycles of generation, evaluation and modification” would have been helpful to enhance model quality and deepen model understanding. Schwarz et al. (2009) suggest a four-step approach to successfully promote progressive understanding of models and scientific modeling. Yet the limited time available in our 1-day outreach laboratory meant that this intervention was not suitable for such extensive modeling activities and we could not offer several cycles of improvement even though this might have positively influenced learning success scores (Louca et al. 2011). Such cycles would instead require regular in-class modeling and evaluation phases (e.g., José et al. 2015). To compensate for the lack of time, a more extensive pre-modeling phase prior to the outreach teaching unit could be included, during which students would be directly introduced to scientific modeling.

Second, our knowledge items about DNA and analytical methods in gene technology laboratories only provide information about the development of students’ factual knowledge. Thus, more open-ended, conceptual questions, such as those in our surveys, would give a deeper insight into student learning.

Conclusion

Although models are already used in biology lessons to encourage scientific reasoning, their effectiveness is rarely scrutinized (Werner et al. 2017). Yet, certain levels of complexity in experimentation and model design are required to adequately support scientific reasoning. Rinehart et al. (2016), therefore, have argued that all cookbook laboratories should be replaced with authentic, epistemic, scientific practices. Maintaining authenticity is, thereby, mandatory, as merely constructing models of scientific phenomena for the sake of modeling would miss the mark. Classroom activities that explicitly introduce students to the nature of models would be far more beneficial (José et al. 2015). Moreover, continuous in-class reflection and discussion are vital for retaining scientific authenticity and familiarizing students with real-life scientific practice (Acevedo 2008). Therefore, we consider our additional evaluation phase to be another valuable approach to integrating real scientific practices into science teaching. Using outreach laboratories, we demonstrated the impact of model-supported teaching on cognitive achievement. Based on our intervention variants, modellers-2 proved to be more effective than modellers-1 when focusing on DNA structure. Our intervention also confirms the effectiveness of research-based laboratory practice and active-learning protocols for cognitive achievement. Every student approaches new learning contents differently, and our gene technology laboratory offers the required flexibility for differentiated teaching, addressing all types and speeds of learning (e.g., Mierdel and Bogner 2019a, b; Chen et al. 2016). Thus, teachers can apply our model evaluation and active-learning approach in the classroom and in other science subjects (e.g., in chemistry education for modeling protein structure (Torres & Correia 2007)). This will, of course, require science teachers to create new materials suitable for complex, inquiry-based lessons. Nonetheless, the modeling has the potential to encourage students to hypothesize, assess the accuracy of explanations, and identify knowledge gaps and is, thus, worth the effort (Svoboda and Passmore 2013). In future studies, extending the modeling phase to include several model-evaluation cycles, as suggested by Hernández et al. (2015), and assessing the impact on actual and persistent learning success would be of interest.